Federal Grant Award Summary

The National Science Foundation's Division of Mathematical Sciences awarded a five-year CAREER grant of $243,000 to Columbia University beginning September 1, 2026, under the Mathematical and Physical Sciences program (CFDA 47.049). This project develops a statistical framework for evaluating uncertainty quantification and principled design of generative artificial intelligence (AI) systems. The research pursues three primary thrusts: (1) measuring overall fidelity of black-box generative models using effective sample size methodology to assess the reliability of synthetic data; (2) extending these methods to provide local, context-aware uncertainty estimates for specific inputs, subpopulations, and dynamic environments; and (3) developing statistically grounded approaches for designing uncertainty-aware generative models.

The project delivers both research outputs and educational/workforce development products. Research deliverables include statistical tools and methodologies that enable organizations to assess when generated data can be trusted, evaluate performance equity across populations, and monitor system performance over time. Practical applications include creation of digital twins of student populations for safe testing of educational technologies. The educational component encompasses K-12 hands-on activities, new graduate-level courses, doctoral student mentoring, and community workshops designed to broaden participation in science and technology. These integrated research and educational products address critical gaps in generative AI reliability assessment while building human capital in statistical AI methods.